def test_change_weights_for_example0b(self):
activation_functions = [ sigmoid_function ]
weight_init_functions = [ self.weight_init_function_test1 ]
learning_rate_functions = [learning_rate_function]
network = self.setup1(activation_functions, weight_init_functions, learning_rate_functions)
single_training_example_inputs = [1.0, 1.0] # includes Bias!
network_outputs = network.calc_networks_output( single_training_example_inputs )
expected_network_output = sigmoid_function(5.0)
assert network_outputs == [expected_network_output]
errors = network.calc_output_neurons_errors(network_outputs, [5.5])
expected_output_error = expected_network_output - 5.5
assert errors == [expected_output_error]
output_neurons_error = network.layers[-1].neurons[0].error
assert output_neurons_error == expected_output_error
output_neurons_error_at_input = network.layers[-1].neurons[0].error_at_input
output_neurons_netinput = network.layers[-1].neurons[0].netinput
assert output_neurons_error_at_input == output_neurons_error * sigmoid_function(output_neurons_netinput, True)
print network.layers[0]
network.change_weights_for_example( single_training_example_inputs )
assert network.layers[0].neurons[0].weights[0] == 2.0 + (output_neurons_error_at_input * 1.0)
assert network.layers[0].neurons[0].weights[1] == 3.0 + (output_neurons_error_at_input * 1.0)
def test_change_weights_for_example0b(self):
activation_functions = [sigmoid_function]
weight_init_functions = [self.weight_init_function_test1]
learning_rate_functions = [learning_rate_function]
network = self.setup1(activation_functions, weight_init_functions,
learning_rate_functions)
single_training_example_inputs = [1.0, 1.0] # includes Bias!
network_outputs = network.calc_networks_output(
single_training_example_inputs)
expected_network_output = sigmoid_function(5.0)
assert network_outputs == [expected_network_output]
errors = network.calc_output_neurons_errors(network_outputs, [5.5])
expected_output_error = expected_network_output - 5.5
assert errors == [expected_output_error]
output_neurons_error = network.layers[-1].neurons[0].error
assert output_neurons_error == expected_output_error
output_neurons_error_at_input = network.layers[-1].neurons[
0].error_at_input
output_neurons_netinput = network.layers[-1].neurons[0].netinput
assert output_neurons_error_at_input == output_neurons_error * sigmoid_function(
output_neurons_netinput, True)
print network.layers[0]
network.change_weights_for_example(single_training_example_inputs)
assert network.layers[0].neurons[0].weights[0] == 2.0 + (
output_neurons_error_at_input * 1.0)
assert network.layers[0].neurons[0].weights[1] == 3.0 + (
output_neurons_error_at_input * 1.0)
def test_calc_hidden_neurons_errors4(self):
activation_functions = [sigmoid_function, sigmoid_function]
weight_init_functions = [
self.weight_init_function_test2, self.weight_init_function_test2
]
learning_rate_functions = [
learning_rate_function, learning_rate_function
]
network = self.setup_single_hidden_layer_net2(activation_functions,
weight_init_functions,
learning_rate_functions)
network_outputs = network.calc_networks_output([1.0, 1.0])
hidden_neuron = network.layers[0].neurons[0]
assert hidden_neuron.netinput == 3.0
expected_hiddens_output = sigmoid_function(3.0)
assert hidden_neuron.output == sigmoid_function(3.0)
output_neuron = network.layers[1].neurons[0]
expected_netinput_to_output_neuron = (2.0 *
expected_hiddens_output) + 1.0
assert output_neuron.netinput == expected_netinput_to_output_neuron
assert network_outputs[0] == sigmoid_function(
expected_netinput_to_output_neuron)
network.calc_output_neurons_errors(network_outputs, [5.5])
expected_output_error = sigmoid_function(
expected_netinput_to_output_neuron) - 5.5
assert output_neuron.error == expected_output_error
assert output_neuron.error_at_input == (
expected_output_error *
sigmoid_function(expected_netinput_to_output_neuron, True))
def setup2(self):
random.seed(1)
aNeuron = Neuron(n_inputs=2, activation_function=sigmoid_function, weight_init_function=weight_init_function_random, learning_rate_function=learning_rate_function)
aNeuron.weights = [1.0,0.5,0.25]
inputs = [2.0,1.5,1.0]
output = aNeuron.calc_neurons_output(inputs)
expected = sigmoid_function( dot(inputs, aNeuron.weights) )
return output, expected
def test_calc_hidden_neurons_errors4(self):
activation_functions = [ sigmoid_function, sigmoid_function ]
weight_init_functions = [ self.weight_init_function_test2, self.weight_init_function_test2 ]
learning_rate_functions = [ learning_rate_function, learning_rate_function ]
network = self.setup_single_hidden_layer_net2(activation_functions, weight_init_functions, learning_rate_functions)
network_outputs = network.calc_networks_output( [1.0, 1.0] )
hidden_neuron = network.layers[0].neurons[0]
assert hidden_neuron.netinput == 3.0
expected_hiddens_output = sigmoid_function(3.0)
assert hidden_neuron.output == sigmoid_function(3.0)
output_neuron = network.layers[1].neurons[0]
expected_netinput_to_output_neuron = (2.0 * expected_hiddens_output) + 1.0
assert output_neuron.netinput == expected_netinput_to_output_neuron
assert network_outputs[0] == sigmoid_function(expected_netinput_to_output_neuron)
network.calc_output_neurons_errors(network_outputs, [5.5])
expected_output_error = sigmoid_function(expected_netinput_to_output_neuron) - 5.5
assert output_neuron.error == expected_output_error
assert output_neuron.error_at_input == (expected_output_error * sigmoid_function( expected_netinput_to_output_neuron, True ))
def setup2(self):
random.seed(1)
aNeuron = Neuron(n_inputs=2,
activation_function=sigmoid_function,
weight_init_function=weight_init_function_random,
learning_rate_function=learning_rate_function)
aNeuron.weights = [1.0, 0.5, 0.25]
inputs = [2.0, 1.5, 1.0]
output = aNeuron.calc_neurons_output(inputs)
expected = sigmoid_function(dot(inputs, aNeuron.weights))
return output, expected
